dhcpcd-discuss

Sounds right - if I add ia_na to my wan0 config, so wan0 gets a global
address directly, the route is added.

I also found out what was tweaking accept_ra to 2 afterwards, and stopped
it.

Nathan

On Mon, Aug 3, 2020 at 5:04 AM Roy Marples <roy@xxxxxxxxxxxx> wrote:

> On 02/08/2020 22:55, Nathan O'Meara wrote:
> > I disabled LAG/LACP, so there's only a real physical lan0 - I'm still
> seeing the
> > same behavior regarding routes.... but now I do see advertisements in
> that
> > tcpdump (and the dhcpcd log), but still no default route added by dhcpcd.
> >
> > However, that matches my earlier logs.. perhaps something else changed
> when I
> > added the bonding option to my kernel.  Since my old logs don't show
> dhcpcd
> > setting a default ipv6 route, my guess is that I was relying on the
> > kernel-native accept_ra setting.  I noticed that for some reason
> > net.ipv6.conf.wan0.accept_ra = 2.  2 seems to be 'always accept router
> > advertisements', but I tried setting it to 1 'accept if forwarding is
> not
> > enabled', and basically immediately the default route was populated.
> >
> > As for not seeing the advertisement packets at all when LAG/LACP was
> enabled,
> > that looks like a firmware bug in my modem.  If the modem side has
> LAG/LACP
> > enabled, but the router side doesn't, ipv4 was working fine, but I still
> didn't
> > see the advertisement packets.  Once I disabled LAG/LACP on the modem,
> the
> > packets started showing up.
>
> Ok .....
>
> > If accept_ra is set to 1 or 2, does dhcpcd not handle the default route
> to avoid
> > two processes handling the same task?
>
> https://roy.marples.name/cgit/dhcpcd.git/tree/src/if-linux.c#n2037
> So when dhcpcd starts IPv6 on an interface, it sets accept_ra to 0 - ie
> disable
> the kernel from handling the RA.
>
> You only want one thing handling the RA - either the kernel, dhcpcd or
> something
> else.
>
> > Aug 02 17:40:43 [6056]: br-3b5335941052: libudev: add
> > Aug 02 17:40:43 [6056]: br-535d72c4732e: libudev: add
> > Aug 02 17:40:43 [6056]: br-185afdecc360: libudev: add
> > Aug 02 17:40:43 [6056]: docker0: libudev: add
> > Aug 02 17:40:43 [6056]: vethd55b05c: libudev: add
> > Aug 02 17:40:43 [6056]: veth262c1b7: libudev: add
> > Aug 02 17:40:43 [6056]: veth9fa4842: libudev: add
> > Aug 02 17:40:43 [6056]: veth70b84de: libudev: add
> > Aug 02 17:40:43 [6056]: veth3bfd070: libudev: add
> > Aug 02 17:40:43 [6056]: veth987f89f: libudev: add
>
> Those are some crazy names!
> Unrelated, just amused me :)
>
> > Aug 02 17:40:46 [6056]: wan0: REPLY6 received from
> fe80::201:5cff:fe66:fe46
> > Aug 02 17:40:46 [6056]: wan0: renew in 302400, rebind in 483840, expire
> in
> > 604800 seconds
> > Aug 02 17:40:46 [6056]: lo: adding reject route to
> 2607:fcc8:9320:1400::/56
> > Aug 02 17:40:46 [6056]: wan0: writing lease `/var/lib/dhcpcd/wan0.lease6'
> > Aug 02 17:40:46 [6056]: wan0: delegated prefix 2607:fcc8:9320:1400::/56
> > Aug 02 17:40:46 [6056]: lan0: adding address 2607:fcc8:9320:1400::1/64
> > Aug 02 17:40:46 [6056]: lan0: pltime 604800 seconds, vltime 604800
> seconds
> > Aug 02 17:40:46 [6056]: lan0: waiting for DHCPv6 DAD to complete
> > Aug 02 17:40:46 [6056]: lan0: adding route to 2607:fcc8:9320:1400::/64
> > Aug 02 17:40:46 [6056]: wan0: executing `/lib/dhcpcd/dhcpcd-run-hooks'
> REBIND6
> > Aug 02 17:40:46 [6056]: waiting for an IP address
> > Aug 02 17:40:46 [6056]: wan0: ARP probing 104.230.3.34 (3 of 3), next in
> 2.0 seconds
> > Aug 02 17:40:47 [6056]: wan0: Router Advertisement from
> fe80::201:5cff:fe66:fe46
> > Aug 02 17:40:47 [6056]: wan0: no global addresses for default route
> > Aug 02 17:40:47 [6056]: wan0: executing `/lib/dhcpcd/dhcpcd-run-hooks'
> ROUTERADVERT
> > Aug 02 17:40:47 [6056]: lan0: DHCPv6 DAD completed
> > Aug 02 17:40:47 [6056]: lan0: executing `/lib/dhcpcd/dhcpcd-run-hooks'
> DELEGATED6
>
> I trimmed your log to just wan0 and lan0.
> It seems that the test for global addresses is not taking into account
> delegated
> addresses and this is causing the default route from the RA not being
> added.
> This looks most definitely like a dhcpcd bug, I'll try and look into it
> later today.
>
> Roy
>